After calibration to unrelated sounds, speech patterns had greater pair agreement than other voices and animal sounds in human iEEG and two fMRI cohorts

We tested whether speech gives more consistent spatial response directions across clips than other human voices and animal sounds. Separate public human studies supplied signed iEEG and BOLD fMRI responses. The locked electrical code pooled three early windows. A later audit found that the prose plan specified only 100–400 ms. That window is a separate post-result check. Unrelated sounds set each site's mean and scale. We then divided each target response vector by its length. Mean cosine over distinct clip pairs measured concentration. This assay is invariant to fixed nonzero affine site changes, including polarity. Four people per modality supplied development. Fixed tests used 20 other patients and 16 other fMRI adults. Mean speech-minus-human and speech-minus-animal scores were 0.32137 and 0.50523 in iEEG, and 0.56906 and 0.55322 in fMRI. All four person mean intervals were positive. Both shared tests passed person mean, sign, and common-label requirements after correction for two claims. The label reference accounted for three planned response definitions. A later fixed follow-up used eight adults from a separate fMRI study. Both scores were positive in all eight. Requiring evidence in all three cohorts gave Holm p=0.0078125 for each claim. The fMRI cohorts used the same clips. Rank and spatial mean controls retained positive reserved means. Acoustic features also had this order. An electrical acoustic residual model failed under extrapolation. Speech had higher fMRI repeat agreement. Models with noise or class mean changes gave positive concentration differences without less latent variation. The result supports measured directional alignment for two fixed panels. It does not establish an abstract speech code, neural compression, a common cause, or causality. The package gives source pins and terms, permitted inputs, plans, all outcomes, figures, independent audits, code, and repeat steps.

auditory cortex · speech · intracranial EEG · fMRI · response geometry · directional concentration · data reuse